Click here to

Session: Therapy General ePoster Viewing [Return to Session]

A Radiomics-Based Light Gradient Boosting Machine to Predict Radiation-Induced Toxicities in Nasopharynx Cancer Patients Receiving Chemoradiotherapy

Z Jiang1*, Y Liang2, X Wang3, Z Min4, M Feng5, Y Kuang6, (1) University of Nevada, Las Vegas, Las Vegas, NV, (2) Cancer Hospital Chinese Academy of Medical Sciences, Sichuan Center, Chengdu, CN, (3) Radiation Oncology Key Laboratory Of Sichuan Province, Chengdu, CN,(4) Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, Chengdu, CN,(5) Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, Chengdu, CN,(6) University of Nevada, Las Vegas, Las Vegas, NV


PO-GePV-T-120 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Oral mucositis, radiodermatitis, skin ulcer, change of sternocleidomastoid muscle thickness and thyrotoxicity are the major radiation-induced toxicities in nasopharynx cancer patients receiving chemoradiotherapy, which significantly affect patients’ quality of life. To predict the increased risk of severe toxicities is still challenging. In this study, we developed a Light Gradient Boosting Machine (Light-GBM) tool to integrate radiomic features within 17 different regions of interest (ROIs) in treatment planning computed tomography (CT) images with clinical features for early prediction of severe radiation-induced toxicities.

Methods: A total of 223 patients with nasopharynx cancer receiving RT were included in this study. The patients were randomly partitioned into training (n=203) and validation (n=20) groups for model development. Clinical features include gender, age, cancer stage, tumor size, tumor location, chemotherapy drugs, distant metastases, radiation therapy position and radiotherapy dose. A total of 756 radiomic features were extracted from GTV, CTV, PTV and OARs regions in images. Toxicities and non-toxicities group classes were labeled within the patients based on the cut-off value of CTCAE ≥ 2. Light-GBM machine was used to develop the model including clinical features only (Model A) and the model combining radiomic and clinical features (Model B). The utility of the constructed models in predicting severe toxicities was evaluated.

Results: For Model A, the AUC values of six toxicities are 0.8, 0.71, 0.72, 0.68, 0.75, and 0.64, respectively. For Model B, the AUC values increased to 0.86, 0.81, 0.84, 0.77, 0.89 and 0.8. The feature important analysis showed that T stage, age, radiation dose, chemotherapy drugs and 14 radiomics features were the most valuable risk predicting factors.

Conclusion: The developed models facilitate identification of patients who are likely to develop severe radiation-induced toxicities, which have a personalized therapeutic implication for appropriate personalized treatment and/or early intervention in the care of nasopharynx cancer patients.


Feature Selection, Statistical Analysis, Modeling


IM- Dataset Analysis/Biomathematics: Machine learning

Contact Email